基于图像处理的尿沉渣图像自动识别研究

发布时间：2018-06-11 11:29

本文选题：图像分割 + 特征提取　；参考：《西安建筑科技大学》2017年硕士论文

【摘要】：目前全自动尿沉渣检测仪是医院实现尿沉渣自动检测的主要手段之一,其识别率的可信性以及准确率对于尿沉渣的自动检测来说是非常重要的。本文展开了对尿沉渣图像自动识别算法的研究,广泛引入了国内外已有的有益成果,并且结合最新发展的技术以及尿沉渣图像的特点,从图像滤波和增强、特征提取和选择、图像分割以及图像识别等几个方面对尿沉渣图像识别进行论述。在整个课题的研究中,在预处理时,首先采用了高斯滤波对图像进行了预处理,然后使用gamma变换对图像进行增强,从而使得图像平滑并且前景和背景对比度变大。在图像分割中,首先使用OSTU方法对图像进行预分割,分割出前景和少量背景,接下来使用异常检测的方法建立模型,将分割图像中的杂质进行去除,与传统方法相比,异常检测方法可以得到分割图片的定量分析结果,使分割效果更加有说服力。在特征提取中,主要提取了细胞的形态学特征,与其他特征相比,形态学特征更加的具有物理意义,然后使用主成分分析算法对特征进行选择。在细胞分类中,本文使用机器学习算法进行分类,通过对比多种机器学习算法并结合尿沉渣图像识别的背景,最终采用支持向量机的分类方法对细胞进行分类。论文基于MATLAB平台以及LIBSVM公开库对采集到的多幅尿沉渣图像运用所研究的算法进行识别检测试验,统计测试结果后表明,本文所研究的算法不仅对尿沉渣图像中的杂质进行准确的过滤,并且能够对图像中的红细胞、白细胞进行比较高效的区分,最终分类准确率达到了95%。
[Abstract]:At present, the automatic urine sediment detector is one of the main means to realize the automatic detection of urinary sediment in hospital. The credibility and accuracy of its recognition rate is very important for the automatic detection of urinary sediment. In this paper, the research on the automatic recognition algorithm of urinary sediment image is carried out, and some useful achievements have been widely introduced, and combined with the latest development technology and the characteristics of urinary sediment image, image filtering and enhancement, feature extraction and selection are introduced. Image segmentation and image recognition are discussed in this paper. In the research of the whole subject, the Gao Si filter is first used to preprocess the image, and then the gamma transform is used to enhance the image, which makes the image smooth and the foreground and background contrast bigger. In image segmentation, OSTU method is first used to segment the image, and the foreground and a few background are segmented. Then, the model is established by using anomaly detection method, and the impurity in the segmented image is removed, which is compared with the traditional method. The method of anomaly detection can get the quantitative analysis result of segmented image, which makes the segmentation effect more persuasive. In feature extraction, the morphological features of cells are mainly extracted. Compared with other features, morphological features have more physical significance, and then the principal component analysis algorithm is used to select the features. In cell classification, machine learning algorithm is used to classify cells. By comparing various machine learning algorithms and combining with the background of urine sediment image recognition, support vector machine is used to classify cells. Based on MATLAB platform and LIBSVM open library, the paper uses the algorithm to identify and detect the collected multiple urine sediment images. The statistical test results show that. The algorithm studied in this paper not only filters the impurities in the urine sediment image accurately, but also can distinguish the red blood cells and white blood cells in the image more efficiently, and finally the classification accuracy reaches 9595%.
【学位授予单位】：西安建筑科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R446.12;TP391.41

【相似文献】